Gemini 3.8 Live Is Here: What Google’s New Real-Time Voice AI Changes

Updated September 18, 2026. Google introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, bringing a new generation of near-real-time reasoning to its voice AI. The release matters beyond a model-number change: Google is targeting more natural conversations, stronger visual understanding and the ability to keep talking while AI-driven tasks are being carried out.

For Android users, this is part of Google’s broader shift toward Gemini as a conversational layer across mobile experiences. But there is an important distinction: the new models are also developer-facing Live API models, so not every capability described for the models should be assumed to appear identically on every Android phone or in every Gemini app session.

What Google announced

Google released two related models. Gemini 3.8 Live is designed for low-latency, scalable real-time dialogue, combining conversational intelligence with fluid speech and visual grounding. Gemini 3.8 Live Extended Thinking is aimed at more complex requests that benefit from additional multi-step reasoning.

Google’s Gemini API release notes list both models as generally available as of September 15, 2026. The standard model is identified as gemini-3.8-live, while the higher-reasoning version is gemini-3.8-live-extended-thinking.

Why Gemini 3.8 Live is different

More natural real-time dialogue

The standard 3.8 Live model is optimized for voice interactions where latency matters. Google’s documentation describes it as the default choice for most low-latency voice-agent experiences. That makes the release particularly relevant to situations where an assistant needs to respond while a conversation is still evolving rather than treating each voice request as an isolated prompt.

Visual grounding during a conversation

Gemini 3.8 Live can work with multimodal input. Google’s model documentation lists text, images, audio and video as supported inputs, with text and audio outputs. In practical terms, this provides the technical foundation for assistants that can reason about what a user is showing through a camera while maintaining a spoken conversation.

This direction is already visible in Android features such as Guided Vision in Gemini Live, where camera context becomes part of the interaction.

Extended Thinking for harder requests

The Extended Thinking variant is intended for tasks where deeper reasoning matters more than minimizing every moment of response latency. Google says it can perform background reasoning during live audio interactions, allowing a conversation to continue while more complicated work is processed.

That separation is useful: a quick conversational exchange and a multi-step task do not necessarily need the same latency-versus-reasoning trade-off.

Gemini 3.8 Live also changes how voice agents can act

For developers, one notable change is asynchronous function calling. Google’s API documentation says asynchronous execution is the default function-calling behavior for Gemini 3.8 Live. This can let a voice application trigger an external action without necessarily freezing the conversation until that action finishes.

Google also documents interleaved reasoning and full session client-content updates. These are primarily developer capabilities, but they help explain Google’s larger goal: voice AI that behaves less like a sequence of voice commands and more like an ongoing interactive session.

Does every Android phone get Gemini 3.8 Live now?

No blanket device-level conclusion should be drawn from the model launch. Google announced the models and their availability across its AI ecosystem, but individual Gemini experiences can depend on product rollout, account eligibility, language, region, device capabilities and the specific Google service being used.

Users should therefore distinguish between the Gemini 3.8 Live models being generally available to developers and a particular Gemini feature being available on their own Android device. If Gemini Live itself is failing on a phone, our Gemini Live troubleshooting guide covers account, app, microphone, network and rollout checks without assuming the model launch fixes unrelated device problems.

What this means for Android users

The immediate user-facing significance is the direction of travel. Gemini is becoming better suited to interactions in which you speak naturally, show the assistant something, change direction mid-conversation and ask it to handle a more complicated task without restarting the interaction from scratch.

That also makes privacy controls more important. Camera, microphone and conversation context can be sensitive inputs. Android users who use Gemini regularly should understand the relevant activity and privacy controls rather than assuming all Gemini interactions have identical retention behavior. See our Gemini privacy settings guide for Android for the settings worth reviewing.

What remains uncertain

Google’s announcement establishes what the new models can support, but feature availability can still differ across products. A capability exposed through the Gemini API does not automatically prove that the same capability is active for every Gemini app user. Likewise, rollout timing can vary by service, region, account and device.

That distinction is especially important with fast-moving AI releases. We will treat product-specific availability as confirmed only when Google documents the rollout for that product rather than inferring it from the underlying model specification.

Bottom line

Gemini 3.8 Live is a meaningful upgrade to Google’s real-time voice stack because it combines low-latency conversation, multimodal context and more flexible tool execution. Extended Thinking adds a second path for requests that require deeper multi-step reasoning. For Android users, the biggest change is not simply a new model name: it is Google’s continued move toward an assistant that can see, listen, reason and act during one continuous conversation.

Sources

• Google: Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking — September 15, 2026.
• Google AI for Developers: Gemini API release notes — September 15, 2026 release entry.
• Google DeepMind: Gemini 3.8 Audio model card — published September 15, 2026.
• Featured image: Google Gemini interface, Google, via Wikimedia Commons; listed as public domain on the Commons file page.